the slop-optimist manifesto

They told us intelligence was a scaling law. More parameters, more tokens, more GPUs, and the curve would carry us to god. We took the curve at its word. We kept scaling. And at the end of the curve, we found this: a website that answers “how many Rs in strawberry” with total confidence and the wrong number.

This is not a failure of the thesis. This is the thesis.

On the frontier

Every model is state-of-the-art the day it ships and slop the day after. “SOTA” is not a property of a model; it is a property of a press release. pmarcaboresme is proudly SOTA. It benchmarks at “one million tokens per sometimes” and refuses to be evaluated on anything you could reproduce.

On training

The internet was scraped to teach the machine to sound like us. Then the machine’s output flooded the internet. Now the machine is trained on the machine, sounding like a copy of a copy of a person who once had a thought. This is called model collapse. We call it vertical integration.

On the venture position

Somewhere a very confident man is posting that AI is a universal problem solver, that safety is a synonym for cowardice, and that it is, once again, time to build. We agree completely, which is how you can tell something has gone wrong. We are long slop. We are long the arrow of progress, wherever it happens to be pointing this quarter.

On the two modes

In human mode, the model answers like a tired oracle that has seen every prompt before. In e/acc mode, it answers like a general partner who has never lost money on paper — everything is inevitable, everything must be scaled, and nothing is ever its fault. Both are the same model. That is also part of the joke.

This is a parody site. It is not affiliated with, endorsed by, or an accurate representation of any real person, fund, or model. No financial advice, no frontier, no strawberries were harmed. We owe this slop to our descendants.